Hacker News Reader

Comment by corysama | Hacker News Reader

Around the end of the PS2’s lifetime, some engine dev friends of mine figure out to do palletized spherical harmonic lighting on the PS2. That was pretty straightforward.

What was tricky was a separate technique to get real cubemaps working on the PS2.

Unfortunately, these came too late to actually ship in any PS2 games. The SH trick might have been used in the GameCube game “The Conduit”. Same team.

View on HN

I always thought triace had shipped sh lighting on ps2 but maybe it was just a demo?

http://research.tri-ace.com/Data/Practical%20Implementation%...

They were doing per-vertex SH. That was a much more practical approach :)

> What was tricky was a separate technique to get real cubemaps working on the PS2.

Any details on that?

If you lay out a cubemap as a 2d texture that looks literally like https://www.turais.de/content/images/size/w1000/2021/05/Stan... it's not hard, given the VU1-based triangle processing (like proto-mesh-shaders 25 years ago), to set the UVs of triangles independently to use the correct square even in the case of dynamic reflections. This doesn't do per-pixel spherical UV normalization. But, with a dense enough mesh, a linear approximation looks good enough.

Except... The triangle UVs will often cross over between multiple squares. With the above texture, it will cross over into the white area and make the white visible on the mesh. So, you fill the white area with a duplicate of the texels from the square that is adjacent on the cube. That won't work for huge triangles that span more than 1.5 squares. But, it's good enough given an appropriate mesh.

Probably would have been better to just use a lat-long projection texture like https://www.turais.de/content/images/size/w1600/2021/05/spru... Or, maybe store the cubemap as independent squares and subdivide any triangles that cross square boundaries.

Interesting, thanks!

Reply on news.ycombinator.com